NEW YORK – Researchers have developed a blood-based test for predicting whether non-Hodgkin lymphoma patients will respond to autologous CAR T-cell therapy prior to starting treatment.
The researchers published results from a training and validation study of the predictive tool, called InflaMix, in Nature Medicine earlier this month. The tool uses machine learning to analyze 14 blood biomarkers associated with inflammation that signal a high tumor burden and is designed to predict whether a patient will respond poorly and have a shorter survival on CD19-directed CAR T-cell therapies.
"This research was inspired by a simple clinical observation that patients with high levels of inflammation before CAR T-cell therapy experienced worse clinical outcomes," said Marcel van den Brink, president of City of Hope Los Angeles and City of Hope National Medical Center, and a senior author of the InflaMix study. "Despite this observation, individual blood tests did not provide a high degree of predictive accuracy."
Autologous CD19-directed CAR T-cell therapies are approved for late-line treatment of non-Hodgkin lymphomas such as follicular lymphoma, mantle cell lymphoma, and B-cell lymphomas. In the US and Europe, Gilead Sciences' Yescarta (axicabtagene ciloleucel), Novartis' Kymriah (tisagenlecleucel), and Bristol Myers Squibb's Breyanzi (lisocabtagene maraleucel) are approved treatments for these diseases. While up to 90 percent of patients can have a complete response to these therapies, about half of those with non-Hodgkin lymphoma will relapse, the study authors estimated in the paper. These patients may also experience severe side effects from the CAR T-cell therapies, such as cytokine release syndrome or neurotoxicity.
The InflaMix signature clusters patients into inflammatory and noninflammatory groups, and the patients in the inflammatory group are less likely to achieve a complete response and more likely to have worse survival outcomes. The researchers used routine blood test results from 149 patients with large B-cell lymphoma (LBCL) treated at Memorial Sloan Kettering Cancer Center in New York to develop the biomarker signature and machine-learning model underlying InflaMix.
These routine tests measured patients' end-organ function, tumor burden, and inflammation. The 14 markers that made up the final signature included markers of organ function such as ALP, AST, albumin, total bilirubin, white blood cell, hemoglobin, and platelet count, along with LDH levels to determine tumor burden; and several markers of inflammation including CRP, ferritin, D-dimer, IL-6, IL-10, and tumor necrosis factor alpha.
Van den Brink noted that these standard blood tests can be performed at any medical center and require only a single blood draw. "While PET scans, molecular tests, and clinical factors can also be informative of response to CAR T, InflaMix has an advantage in that it combines its predictive strength with broad applicability," he said.
After the model was developed, the researchers conducted further analyses to evaluate its predictive performance. When adjusted for clinical features, the LBCL patients assigned to the inflammatory group by InflaMix had increased odds of not achieving a complete response by day 100, reduced progression-free survival, and reduced overall survival.
In a secondary analysis, researchers also explored InflaMix's performance in separate cohorts of patients with follicular lymphoma and mantle cell lymphoma treated at MSK, Sheba Medical Center in Israel, and Hackensack Meridian Health in New Jersey. They found that InflaMix-assigned inflammatory groups comprising patients with these types of lymphomas also had lower complete response rates and shorter progression-free survival. In total, the researchers studied InflaMix in three independent cohorts comprising 688 patients with non-Hodgkin lymphoma.
"We showed that InflaMix was well calibrated and more discriminating than alternative biomarkers in lymphoma," van den Brink said. "Importantly, using a statistical method called decision curve analysis, we showed that InflaMix also improves benefit for clinical decision-making about whether a patient would benefit from additional treatment after their CAR T treatment."
In the decision curve analysis, the researchers pitted InflaMix against other biomarkers in a hypothetical situation where a physician may have to decide whether to pursue further treatment with consolidation therapy with bispecific T-cell engager therapy or give autologous hematopoietic cell transplantation to patients achieving early partial response one month after CAR T-cell therapy. They found that InflaMix performed statistically better than using other clinical and biomarker models at predicting progression-free survival at six months. This means InflaMix is more informative to physicians who may be considering whether to give additional treatment to their patients receiving a CAR T-cell therapy, according to the researchers.
Since it is important to van den Brink and colleagues that InflaMix can be performed at any medical center, the researchers also evaluated how the model performed with fewer biomarkers. Using a version of the model that included only six of the 14 biomarkers — albumin, Hgb, AST, ALP, CRP and LDH — the test remained reliable. Patients assigned to the inflammatory group by the six-biomarker model also had worse response and survival than those in the noninflammatory group.
While InflaMix is only available in the research setting, van den Brink said his team aims to begin a clinical trial to prospectively evaluate the model and bring it closer to clinical use. His team also hopes to conduct further research on the role of inflammation in poor CAR T-cell therapy response and to evaluate whether InflaMix can predict response or outcomes in other immunotherapy contexts.
"Our goal is to eventually have InflaMix used for identifying patients who would benefit from additional treatment, either anti-inflammatory drugs before CAR T or additional anti-lymphoma drugs after CAR T," he said.